Pipeline-level differentiable programming for the real world
summary
Presents an open-source system that wraps Python modules as Tessract Core containers exposing a containerized REST API and auto-generates Pydantic schemas, Jacobian/JVP endpoints and CLI docs; registers calls as a JAX primitive so JAX.grad traces through HTTP-backed components; uses context managers to launch containers on demand; handles non-differentiable geometry with finite differences and swaps OpenFOAM for learned surrogates in an optimization loop—the framework is usable and traces gradients end-to-end.
AI-written summary · about
chapters (8)
- 0:00Introduction to differentiable programming
- 3:42Automatic differentiation basics
- 6:51Engineering challenges in pipelines
- 11:45Tesseracts framework design
- 17:03JAX integration
- 19:34Realistic pipeline implementation
- 21:56Conclusion and Q&A
- 23:39Technical Q&A deep dive
related talks
01h 10mThe impact of differentiable programming: how ∂P is enabling new science in Julia
54mGeneralized Physics-Informed Learning through Language-Wide Differentiable Programming
11mIntro to JAX: Accelerating Machine Learning research
01h 01mProgramming Abstractions for Orchestration of HPC Scientific Computing
32mProbabilistic and Differentiable Programming in Scientific Simulators
29mAutomatic differentiation, a tale of two languages
59mDifferentiable Programming in C++
01h 10mJAX: accelerated machine learning research via composable function transformations in Python